In [2]:
import os
import tensorflow as tf
os.environ['TFHUB_MODEL_LOAD_FORMAT'] = 'COMPRESSED'
In [3]:
import IPython.display as display

import matplotlib.pyplot as plt
import matplotlib as mpl
mpl.rcParams['figure.figsize'] = (12, 12)
mpl.rcParams['axes.grid'] = False

import numpy as np
import PIL.Image
import time
import functools
In [4]:
def tensor_to_image(tensor):
  tensor = tensor*255
  tensor = np.array(tensor, dtype=np.uint8)
  if np.ndim(tensor)>3:
    assert tensor.shape[0] == 1
    tensor = tensor[0]
  return PIL.Image.fromarray(tensor)
In [5]:
def load_img(path_to_img):
  max_dim = 512
  img = tf.io.read_file(path_to_img)
  img = tf.image.decode_image(img, channels=3)
  img = tf.image.convert_image_dtype(img, tf.float32)

  shape = tf.cast(tf.shape(img)[:-1], tf.float32)
  long_dim = max(shape)
  scale = max_dim / long_dim

  new_shape = tf.cast(shape * scale, tf.int32)

  img = tf.image.resize(img, new_shape)
  img = img[tf.newaxis, :]
  return img
In [6]:
def imshow(image, title=None):
  if len(image.shape) > 3:
    image = tf.squeeze(image, axis=0)

  plt.imshow(image)
  if title:
    plt.title(title)
In [7]:
content_image = load_img("owl.jpg")
style_image = load_img("style.jpeg")

plt.subplot(1, 2, 1)
imshow(content_image, 'Content Image')

plt.subplot(1, 2, 2)
imshow(style_image, 'Style Image')
In [8]:
import tensorflow_hub as hub
hub_model = hub.load('https://tfhub.dev/google/magenta/arbitrary-image-stylization-v1-256/2')
stylized_image = hub_model(tf.constant(content_image), tf.constant(style_image))[0]
tensor_to_image(stylized_image)
Out[8]:
In [9]:
x = tf.keras.applications.vgg19.preprocess_input(content_image*255)
x = tf.image.resize(x, (224, 224))
vgg = tf.keras.applications.VGG19(include_top=True, weights='imagenet')
prediction_probabilities = vgg(x)
prediction_probabilities.shape

Downloading data from https://storage.googleapis.com/tensorflow/keras-applications/vgg19/vgg19_weights_tf_dim_ordering_tf_kernels.h5
574717952/574710816 [==============================] - 4s 0us/step
Out[9]:
TensorShape([1, 1000])
In [10]:
predicted_top_5 = tf.keras.applications.vgg19.decode_predictions(prediction_probabilities.numpy())[0]
[(class_name, prob) for (number, class_name, prob) in predicted_top_5]

Downloading data from https://storage.googleapis.com/download.tensorflow.org/data/imagenet_class_index.json
40960/35363 [==================================] - 0s 0us/step
Out[10]:
[('great_grey_owl', 0.73142934),
 ('ringlet', 0.051973876),
 ('bittern', 0.049039505),
 ('vulture', 0.03952588),
 ('kite', 0.029020881)]
In [21]:
vgg = tf.keras.applications.VGG19(include_top=False, weights='imagenet')

print()
for layer in vgg.layers:
  print(layer.name)

input_6
block1_conv1
block1_conv2
block1_pool
block2_conv1
block2_conv2
block2_pool
block3_conv1
block3_conv2
block3_conv3
block3_conv4
block3_pool
block4_conv1
block4_conv2
block4_conv3
block4_conv4
block4_pool
block5_conv1
block5_conv2
block5_conv3
block5_conv4
block5_pool
In [23]:
content_layers = ['block5_conv2'] 

style_layers = ['block1_conv1',
                'block2_conv1',
                'block3_conv1', 
                'block4_conv1', 
                'block5_conv1']

num_content_layers = len(content_layers)
num_style_layers = len(style_layers)
In [13]:
def vgg_layers(layer_names):
 
  vgg = tf.keras.applications.VGG19(include_top=False, weights='imagenet')
  vgg.trainable = False
  
  outputs = [vgg.get_layer(name).output for name in layer_names]

  model = tf.keras.Model([vgg.input], outputs)
  return model
In [24]:
style_extractor = vgg_layers(style_layers)
style_outputs = style_extractor(style_image*255)


for name, output in zip(style_layers, style_outputs):
  print(name)
  print("  shape: ", output.numpy().shape)
  print("  min: ", output.numpy().min())
  print("  max: ", output.numpy().max())
  print("  mean: ", output.numpy().mean())
  print()

block1_conv1
  shape:  (1, 341, 512, 64)
  min:  0.0
  max:  836.31885
  mean:  49.30245

block2_conv1
  shape:  (1, 170, 256, 128)
  min:  0.0
  max:  5778.871
  mean:  260.07843

block3_conv1
  shape:  (1, 85, 128, 256)
  min:  0.0
  max:  12648.359
  mean:  290.02325

block4_conv1
  shape:  (1, 42, 64, 512)
  min:  0.0
  max:  25104.59
  mean:  917.1499

block5_conv1
  shape:  (1, 21, 32, 512)
  min:  0.0
  max:  3712.8596
  mean:  59.95148
In [25]:
def gram_matrix(input_tensor):
  result = tf.linalg.einsum('bijc,bijd->bcd', input_tensor, input_tensor)
  input_shape = tf.shape(input_tensor)
  num_locations = tf.cast(input_shape[1]*input_shape[2], tf.float32)
  return result/(num_locations)
In [26]:
class StyleContentModel(tf.keras.models.Model):
  def __init__(self, style_layers, content_layers):
    super(StyleContentModel, self).__init__()
    self.vgg = vgg_layers(style_layers + content_layers)
    self.style_layers = style_layers
    self.content_layers = content_layers
    self.num_style_layers = len(style_layers)
    self.vgg.trainable = False

  def call(self, inputs):
    "Expects float input in [0,1]"
    inputs = inputs*255.0
    preprocessed_input = tf.keras.applications.vgg19.preprocess_input(inputs)
    outputs = self.vgg(preprocessed_input)
    style_outputs, content_outputs = (outputs[:self.num_style_layers],
                                      outputs[self.num_style_layers:])

    style_outputs = [gram_matrix(style_output)
                     for style_output in style_outputs]

    content_dict = {content_name: value
                    for content_name, value
                    in zip(self.content_layers, content_outputs)}

    style_dict = {style_name: value
                  for style_name, value
                  in zip(self.style_layers, style_outputs)}

    return {'content': content_dict, 'style': style_dict}
In [27]:
extractor = StyleContentModel(style_layers, content_layers)

results = extractor(tf.constant(content_image))

print('Styles:')
for name, output in sorted(results['style'].items()):
  print("  ", name)
  print("    shape: ", output.numpy().shape)
  print("    min: ", output.numpy().min())
  print("    max: ", output.numpy().max())
  print("    mean: ", output.numpy().mean())
  print()

print("Contents:")
for name, output in sorted(results['content'].items()):
  print("  ", name)
  print("    shape: ", output.numpy().shape)
  print("    min: ", output.numpy().min())
  print("    max: ", output.numpy().max())
  print("    mean: ", output.numpy().mean())

Styles:
   block1_conv1
    shape:  (1, 64, 64)
    min:  0.002655763
    max:  16853.451
    mean:  337.22418

   block2_conv1
    shape:  (1, 128, 128)
    min:  0.0
    max:  48416.816
    mean:  9061.926

   block3_conv1
    shape:  (1, 256, 256)
    min:  0.0
    max:  317588.75
    mean:  8140.998

   block4_conv1
    shape:  (1, 512, 512)
    min:  0.0
    max:  3212877.0
    mean:  145063.1

   block5_conv1
    shape:  (1, 512, 512)
    min:  0.0
    max:  129496.65
    mean:  1231.7739

Contents:
   block5_conv2
    shape:  (1, 22, 32, 512)
    min:  0.0
    max:  1756.408
    mean:  12.337271
In [28]:
style_targets = extractor(style_image)['style']
content_targets = extractor(content_image)['content']
In [29]:
image = tf.Variable(content_image)
In [30]:
def clip_0_1(image):
  return tf.clip_by_value(image, clip_value_min=0.0, clip_value_max=1.0)
In [31]:
opt = tf.optimizers.Adam(learning_rate=0.02, beta_1=0.99, epsilon=1e-1)
In [32]:
style_weight=1e-2
content_weight=1e4
In [33]:
def style_content_loss(outputs):
    style_outputs = outputs['style']
    content_outputs = outputs['content']
    style_loss = tf.add_n([tf.reduce_mean((style_outputs[name]-style_targets[name])**2) 
                           for name in style_outputs.keys()])
    style_loss *= style_weight / num_style_layers

    content_loss = tf.add_n([tf.reduce_mean((content_outputs[name]-content_targets[name])**2) 
                             for name in content_outputs.keys()])
    content_loss *= content_weight / num_content_layers
    loss = style_loss + content_loss
    return loss
In [34]:
@tf.function()
def train_step(image):
  with tf.GradientTape() as tape:
    outputs = extractor(image)
    loss = style_content_loss(outputs)

  grad = tape.gradient(loss, image)
  opt.apply_gradients([(grad, image)])
  image.assign(clip_0_1(image))
In [35]:
train_step(image)
train_step(image)
train_step(image)
tensor_to_image(image)
Out[35]:
In [36]:
import time
start = time.time()

epochs = 10
steps_per_epoch = 100

step = 0
for n in range(epochs):
  for m in range(steps_per_epoch):
    step += 1
    train_step(image)
    print(".", end='', flush=True)
  display.clear_output(wait=True)
  display.display(tensor_to_image(image))
  print("Train step: {}".format(step))
  
end = time.time()
print("Total time: {:.1f}".format(end-start))

Train step: 1000
Total time: 60.6
In [37]:
def high_pass_x_y(image):
  x_var = image[:, :, 1:, :] - image[:, :, :-1, :]
  y_var = image[:, 1:, :, :] - image[:, :-1, :, :]

  return x_var, y_var
In [38]:
x_deltas, y_deltas = high_pass_x_y(content_image)

plt.figure(figsize=(14, 10))
plt.subplot(2, 2, 1)
imshow(clip_0_1(2*y_deltas+0.5), "Horizontal Deltas: Original")

plt.subplot(2, 2, 2)
imshow(clip_0_1(2*x_deltas+0.5), "Vertical Deltas: Original")

x_deltas, y_deltas = high_pass_x_y(image)

plt.subplot(2, 2, 3)
imshow(clip_0_1(2*y_deltas+0.5), "Horizontal Deltas: Styled")

plt.subplot(2, 2, 4)
imshow(clip_0_1(2*x_deltas+0.5), "Vertical Deltas: Styled")
In [39]:
plt.figure(figsize=(14, 10))

sobel = tf.image.sobel_edges(content_image)
plt.subplot(1, 2, 1)
imshow(clip_0_1(sobel[..., 0]/4+0.5), "Horizontal Sobel-edges")
plt.subplot(1, 2, 2)
imshow(clip_0_1(sobel[..., 1]/4+0.5), "Vertical Sobel-edges")
In [40]:
def total_variation_loss(image):
  x_deltas, y_deltas = high_pass_x_y(image)
  return tf.reduce_sum(tf.abs(x_deltas)) + tf.reduce_sum(tf.abs(y_deltas))
In [41]:
total_variation_loss(image).numpy()
Out[41]:
144930.08
In [42]:
tf.image.total_variation(image).numpy()
Out[42]:
array([144930.08], dtype=float32)
In [43]:
total_variation_weight=30
In [44]:
@tf.function()
def train_step(image):
  with tf.GradientTape() as tape:
    outputs = extractor(image)
    loss = style_content_loss(outputs)
    loss += total_variation_weight*tf.image.total_variation(image)

  grad = tape.gradient(loss, image)
  opt.apply_gradients([(grad, image)])
  image.assign(clip_0_1(image))
In [45]:
image = tf.Variable(content_image)
In [46]:
import time
start = time.time()

epochs = 10
steps_per_epoch = 100

step = 0
for n in range(epochs):
  for m in range(steps_per_epoch):
    step += 1
    train_step(image)
    print(".", end='', flush=True)
  display.clear_output(wait=True)
  display.display(tensor_to_image(image))
  print("Train step: {}".format(step))

end = time.time()
print("Total time: {:.1f}".format(end-start))

Train step: 1000
Total time: 63.2
In [47]:
file_name = 'stylized_tranfer_image.png'
tensor_to_image(image).save(file_name)

try:
  from google.colab import files
except ImportError:
   pass
else:
  files.download(file_name)